CUE

CUE imputes DNA methylation levels at CpG sites present on the Illumina HumanMethylationEPIC (HM850) BeadChip from data generated on the Illumina HumanMethylation450 (HM450) BeadChip to enable integration of methylation datasets across profiling platforms.

Key Features:

• Hybrid Methodology: Combines classical statistical methods and modern machine learning in an ensemble strategy to predict HM850 CpG methylation from HM450 data.
• Performance Metrics: In validation on the ELGAN cohort and the VA Boston Posttraumatic Stress Disorder (PTSD) genetics repository, CUE outperformed k-nearest-neighbors, logistic regression, penalized functional regression, random forest, and XGBoost, achieving RMSE = 0.026 and accuracy = 99.97% in the PTSD cohort.
• Imputation Success: Successfully imputed 289,604 of 339,033 HM850-unique CpG sites shared between ELGAN and PTSD (85.4%) using thresholds of RMSE < 0.05 and accuracy > 95% in the PTSD cohort.

Scientific Applications:

• Platform harmonization: Enables integrated analyses across studies that use HM450 or HM850 arrays by imputing unmeasured HM850 CpG sites.
• Epidemiological and clinical methylation studies: Supports large-scale comparative and reproducibility analyses in epidemiology and clinical research requiring comprehensive CpG coverage.

Methodology:

Ensembles outputs from multiple algorithms—k-nearest-neighbors, logistic regression, penalized functional regression, random forest, and XGBoost—to predict HM850 CpG methylation from HM450 data and assesses imputation success using RMSE and accuracy thresholds.